1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * S390 version 4 * Copyright IBM Corp. 1999 5 * Author(s): Hartmut Penner (hp@de.ibm.com) 6 * Ulrich Weigand (uweigand@de.ibm.com) 7 * 8 * Derived from "arch/i386/mm/fault.c" 9 * Copyright (C) 1995 Linus Torvalds 10 */ 11 12 #include <linux/kernel_stat.h> 13 #include <linux/perf_event.h> 14 #include <linux/signal.h> 15 #include <linux/sched.h> 16 #include <linux/sched/debug.h> 17 #include <linux/kernel.h> 18 #include <linux/errno.h> 19 #include <linux/string.h> 20 #include <linux/types.h> 21 #include <linux/ptrace.h> 22 #include <linux/mman.h> 23 #include <linux/mm.h> 24 #include <linux/compat.h> 25 #include <linux/smp.h> 26 #include <linux/kdebug.h> 27 #include <linux/init.h> 28 #include <linux/console.h> 29 #include <linux/extable.h> 30 #include <linux/hardirq.h> 31 #include <linux/kprobes.h> 32 #include <linux/uaccess.h> 33 #include <linux/hugetlb.h> 34 #include <asm/asm-offsets.h> 35 #include <asm/diag.h> 36 #include <asm/gmap.h> 37 #include <asm/irq.h> 38 #include <asm/mmu_context.h> 39 #include <asm/facility.h> 40 #include <asm/uv.h> 41 #include "../kernel/entry.h" 42 43 #define __FAIL_ADDR_MASK -4096L 44 #define __SUBCODE_MASK 0x0600 45 #define __PF_RES_FIELD 0x8000000000000000ULL 46 47 #define VM_FAULT_BADCONTEXT ((__force vm_fault_t) 0x010000) 48 #define VM_FAULT_BADMAP ((__force vm_fault_t) 0x020000) 49 #define VM_FAULT_BADACCESS ((__force vm_fault_t) 0x040000) 50 #define VM_FAULT_SIGNAL ((__force vm_fault_t) 0x080000) 51 #define VM_FAULT_PFAULT ((__force vm_fault_t) 0x100000) 52 53 enum fault_type { 54 KERNEL_FAULT, 55 USER_FAULT, 56 VDSO_FAULT, 57 GMAP_FAULT, 58 }; 59 60 static unsigned long store_indication __read_mostly; 61 62 static int __init fault_init(void) 63 { 64 if (test_facility(75)) 65 store_indication = 0xc00; 66 return 0; 67 } 68 early_initcall(fault_init); 69 70 /* 71 * Find out which address space caused the exception. 72 */ 73 static enum fault_type get_fault_type(struct pt_regs *regs) 74 { 75 unsigned long trans_exc_code; 76 77 trans_exc_code = regs->int_parm_long & 3; 78 if (likely(trans_exc_code == 0)) { 79 /* primary space exception */ 80 if (IS_ENABLED(CONFIG_PGSTE) && 81 test_pt_regs_flag(regs, PIF_GUEST_FAULT)) 82 return GMAP_FAULT; 83 if (current->thread.mm_segment == USER_DS) 84 return USER_FAULT; 85 return KERNEL_FAULT; 86 } 87 if (trans_exc_code == 2) { 88 /* secondary space exception */ 89 if (current->thread.mm_segment & 1) { 90 if (current->thread.mm_segment == USER_DS_SACF) 91 return USER_FAULT; 92 return KERNEL_FAULT; 93 } 94 return VDSO_FAULT; 95 } 96 if (trans_exc_code == 1) { 97 /* access register mode, not used in the kernel */ 98 return USER_FAULT; 99 } 100 /* home space exception -> access via kernel ASCE */ 101 return KERNEL_FAULT; 102 } 103 104 static int bad_address(void *p) 105 { 106 unsigned long dummy; 107 108 return get_kernel_nofault(dummy, (unsigned long *)p); 109 } 110 111 static void dump_pagetable(unsigned long asce, unsigned long address) 112 { 113 unsigned long *table = __va(asce & _ASCE_ORIGIN); 114 115 pr_alert("AS:%016lx ", asce); 116 switch (asce & _ASCE_TYPE_MASK) { 117 case _ASCE_TYPE_REGION1: 118 table += (address & _REGION1_INDEX) >> _REGION1_SHIFT; 119 if (bad_address(table)) 120 goto bad; 121 pr_cont("R1:%016lx ", *table); 122 if (*table & _REGION_ENTRY_INVALID) 123 goto out; 124 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 125 fallthrough; 126 case _ASCE_TYPE_REGION2: 127 table += (address & _REGION2_INDEX) >> _REGION2_SHIFT; 128 if (bad_address(table)) 129 goto bad; 130 pr_cont("R2:%016lx ", *table); 131 if (*table & _REGION_ENTRY_INVALID) 132 goto out; 133 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 134 fallthrough; 135 case _ASCE_TYPE_REGION3: 136 table += (address & _REGION3_INDEX) >> _REGION3_SHIFT; 137 if (bad_address(table)) 138 goto bad; 139 pr_cont("R3:%016lx ", *table); 140 if (*table & (_REGION_ENTRY_INVALID | _REGION3_ENTRY_LARGE)) 141 goto out; 142 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 143 fallthrough; 144 case _ASCE_TYPE_SEGMENT: 145 table += (address & _SEGMENT_INDEX) >> _SEGMENT_SHIFT; 146 if (bad_address(table)) 147 goto bad; 148 pr_cont("S:%016lx ", *table); 149 if (*table & (_SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_LARGE)) 150 goto out; 151 table = (unsigned long *)(*table & _SEGMENT_ENTRY_ORIGIN); 152 } 153 table += (address & _PAGE_INDEX) >> _PAGE_SHIFT; 154 if (bad_address(table)) 155 goto bad; 156 pr_cont("P:%016lx ", *table); 157 out: 158 pr_cont("\n"); 159 return; 160 bad: 161 pr_cont("BAD\n"); 162 } 163 164 static void dump_fault_info(struct pt_regs *regs) 165 { 166 unsigned long asce; 167 168 pr_alert("Failing address: %016lx TEID: %016lx\n", 169 regs->int_parm_long & __FAIL_ADDR_MASK, regs->int_parm_long); 170 pr_alert("Fault in "); 171 switch (regs->int_parm_long & 3) { 172 case 3: 173 pr_cont("home space "); 174 break; 175 case 2: 176 pr_cont("secondary space "); 177 break; 178 case 1: 179 pr_cont("access register "); 180 break; 181 case 0: 182 pr_cont("primary space "); 183 break; 184 } 185 pr_cont("mode while using "); 186 switch (get_fault_type(regs)) { 187 case USER_FAULT: 188 asce = S390_lowcore.user_asce; 189 pr_cont("user "); 190 break; 191 case VDSO_FAULT: 192 asce = S390_lowcore.vdso_asce; 193 pr_cont("vdso "); 194 break; 195 case GMAP_FAULT: 196 asce = ((struct gmap *) S390_lowcore.gmap)->asce; 197 pr_cont("gmap "); 198 break; 199 case KERNEL_FAULT: 200 asce = S390_lowcore.kernel_asce; 201 pr_cont("kernel "); 202 break; 203 default: 204 unreachable(); 205 } 206 pr_cont("ASCE.\n"); 207 dump_pagetable(asce, regs->int_parm_long & __FAIL_ADDR_MASK); 208 } 209 210 int show_unhandled_signals = 1; 211 212 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault) 213 { 214 if ((task_pid_nr(current) > 1) && !show_unhandled_signals) 215 return; 216 if (!unhandled_signal(current, signr)) 217 return; 218 if (!printk_ratelimit()) 219 return; 220 printk(KERN_ALERT "User process fault: interruption code %04x ilc:%d ", 221 regs->int_code & 0xffff, regs->int_code >> 17); 222 print_vma_addr(KERN_CONT "in ", regs->psw.addr); 223 printk(KERN_CONT "\n"); 224 if (is_mm_fault) 225 dump_fault_info(regs); 226 show_regs(regs); 227 } 228 229 /* 230 * Send SIGSEGV to task. This is an external routine 231 * to keep the stack usage of do_page_fault small. 232 */ 233 static noinline void do_sigsegv(struct pt_regs *regs, int si_code) 234 { 235 report_user_fault(regs, SIGSEGV, 1); 236 force_sig_fault(SIGSEGV, si_code, 237 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK)); 238 } 239 240 const struct exception_table_entry *s390_search_extables(unsigned long addr) 241 { 242 const struct exception_table_entry *fixup; 243 244 fixup = search_extable(__start_dma_ex_table, 245 __stop_dma_ex_table - __start_dma_ex_table, 246 addr); 247 if (!fixup) 248 fixup = search_exception_tables(addr); 249 return fixup; 250 } 251 252 static noinline void do_no_context(struct pt_regs *regs) 253 { 254 const struct exception_table_entry *fixup; 255 256 /* Are we prepared to handle this kernel fault? */ 257 fixup = s390_search_extables(regs->psw.addr); 258 if (fixup) { 259 regs->psw.addr = extable_fixup(fixup); 260 return; 261 } 262 263 /* 264 * Oops. The kernel tried to access some bad page. We'll have to 265 * terminate things with extreme prejudice. 266 */ 267 if (get_fault_type(regs) == KERNEL_FAULT) 268 printk(KERN_ALERT "Unable to handle kernel pointer dereference" 269 " in virtual kernel address space\n"); 270 else 271 printk(KERN_ALERT "Unable to handle kernel paging request" 272 " in virtual user address space\n"); 273 dump_fault_info(regs); 274 die(regs, "Oops"); 275 do_exit(SIGKILL); 276 } 277 278 static noinline void do_low_address(struct pt_regs *regs) 279 { 280 /* Low-address protection hit in kernel mode means 281 NULL pointer write access in kernel mode. */ 282 if (regs->psw.mask & PSW_MASK_PSTATE) { 283 /* Low-address protection hit in user mode 'cannot happen'. */ 284 die (regs, "Low-address protection"); 285 do_exit(SIGKILL); 286 } 287 288 do_no_context(regs); 289 } 290 291 static noinline void do_sigbus(struct pt_regs *regs) 292 { 293 /* 294 * Send a sigbus, regardless of whether we were in kernel 295 * or user mode. 296 */ 297 force_sig_fault(SIGBUS, BUS_ADRERR, 298 (void __user *)(regs->int_parm_long & __FAIL_ADDR_MASK)); 299 } 300 301 static noinline int signal_return(struct pt_regs *regs) 302 { 303 u16 instruction; 304 int rc; 305 306 rc = __get_user(instruction, (u16 __user *) regs->psw.addr); 307 if (rc) 308 return rc; 309 if (instruction == 0x0a77) { 310 set_pt_regs_flag(regs, PIF_SYSCALL); 311 regs->int_code = 0x00040077; 312 return 0; 313 } else if (instruction == 0x0aad) { 314 set_pt_regs_flag(regs, PIF_SYSCALL); 315 regs->int_code = 0x000400ad; 316 return 0; 317 } 318 return -EACCES; 319 } 320 321 static noinline void do_fault_error(struct pt_regs *regs, int access, 322 vm_fault_t fault) 323 { 324 int si_code; 325 326 switch (fault) { 327 case VM_FAULT_BADACCESS: 328 if (access == VM_EXEC && signal_return(regs) == 0) 329 break; 330 fallthrough; 331 case VM_FAULT_BADMAP: 332 /* Bad memory access. Check if it is kernel or user space. */ 333 if (user_mode(regs)) { 334 /* User mode accesses just cause a SIGSEGV */ 335 si_code = (fault == VM_FAULT_BADMAP) ? 336 SEGV_MAPERR : SEGV_ACCERR; 337 do_sigsegv(regs, si_code); 338 break; 339 } 340 fallthrough; 341 case VM_FAULT_BADCONTEXT: 342 case VM_FAULT_PFAULT: 343 do_no_context(regs); 344 break; 345 case VM_FAULT_SIGNAL: 346 if (!user_mode(regs)) 347 do_no_context(regs); 348 break; 349 default: /* fault & VM_FAULT_ERROR */ 350 if (fault & VM_FAULT_OOM) { 351 if (!user_mode(regs)) 352 do_no_context(regs); 353 else 354 pagefault_out_of_memory(); 355 } else if (fault & VM_FAULT_SIGSEGV) { 356 /* Kernel mode? Handle exceptions or die */ 357 if (!user_mode(regs)) 358 do_no_context(regs); 359 else 360 do_sigsegv(regs, SEGV_MAPERR); 361 } else if (fault & VM_FAULT_SIGBUS) { 362 /* Kernel mode? Handle exceptions or die */ 363 if (!user_mode(regs)) 364 do_no_context(regs); 365 else 366 do_sigbus(regs); 367 } else 368 BUG(); 369 break; 370 } 371 } 372 373 /* 374 * This routine handles page faults. It determines the address, 375 * and the problem, and then passes it off to one of the appropriate 376 * routines. 377 * 378 * interruption code (int_code): 379 * 04 Protection -> Write-Protection (suprression) 380 * 10 Segment translation -> Not present (nullification) 381 * 11 Page translation -> Not present (nullification) 382 * 3b Region third trans. -> Not present (nullification) 383 */ 384 static inline vm_fault_t do_exception(struct pt_regs *regs, int access) 385 { 386 struct gmap *gmap; 387 struct task_struct *tsk; 388 struct mm_struct *mm; 389 struct vm_area_struct *vma; 390 enum fault_type type; 391 unsigned long trans_exc_code; 392 unsigned long address; 393 unsigned int flags; 394 vm_fault_t fault; 395 396 tsk = current; 397 /* 398 * The instruction that caused the program check has 399 * been nullified. Don't signal single step via SIGTRAP. 400 */ 401 clear_pt_regs_flag(regs, PIF_PER_TRAP); 402 403 if (kprobe_page_fault(regs, 14)) 404 return 0; 405 406 mm = tsk->mm; 407 trans_exc_code = regs->int_parm_long; 408 409 /* 410 * Verify that the fault happened in user space, that 411 * we are not in an interrupt and that there is a 412 * user context. 413 */ 414 fault = VM_FAULT_BADCONTEXT; 415 type = get_fault_type(regs); 416 switch (type) { 417 case KERNEL_FAULT: 418 goto out; 419 case VDSO_FAULT: 420 fault = VM_FAULT_BADMAP; 421 goto out; 422 case USER_FAULT: 423 case GMAP_FAULT: 424 if (faulthandler_disabled() || !mm) 425 goto out; 426 break; 427 } 428 429 address = trans_exc_code & __FAIL_ADDR_MASK; 430 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 431 flags = FAULT_FLAG_DEFAULT; 432 if (user_mode(regs)) 433 flags |= FAULT_FLAG_USER; 434 if (access == VM_WRITE || (trans_exc_code & store_indication) == 0x400) 435 flags |= FAULT_FLAG_WRITE; 436 mmap_read_lock(mm); 437 438 gmap = NULL; 439 if (IS_ENABLED(CONFIG_PGSTE) && type == GMAP_FAULT) { 440 gmap = (struct gmap *) S390_lowcore.gmap; 441 current->thread.gmap_addr = address; 442 current->thread.gmap_write_flag = !!(flags & FAULT_FLAG_WRITE); 443 current->thread.gmap_int_code = regs->int_code & 0xffff; 444 address = __gmap_translate(gmap, address); 445 if (address == -EFAULT) { 446 fault = VM_FAULT_BADMAP; 447 goto out_up; 448 } 449 if (gmap->pfault_enabled) 450 flags |= FAULT_FLAG_RETRY_NOWAIT; 451 } 452 453 retry: 454 fault = VM_FAULT_BADMAP; 455 vma = find_vma(mm, address); 456 if (!vma) 457 goto out_up; 458 459 if (unlikely(vma->vm_start > address)) { 460 if (!(vma->vm_flags & VM_GROWSDOWN)) 461 goto out_up; 462 if (expand_stack(vma, address)) 463 goto out_up; 464 } 465 466 /* 467 * Ok, we have a good vm_area for this memory access, so 468 * we can handle it.. 469 */ 470 fault = VM_FAULT_BADACCESS; 471 if (unlikely(!(vma->vm_flags & access))) 472 goto out_up; 473 474 if (is_vm_hugetlb_page(vma)) 475 address &= HPAGE_MASK; 476 /* 477 * If for any reason at all we couldn't handle the fault, 478 * make sure we exit gracefully rather than endlessly redo 479 * the fault. 480 */ 481 fault = handle_mm_fault(vma, address, flags); 482 if (fault_signal_pending(fault, regs)) { 483 fault = VM_FAULT_SIGNAL; 484 if (flags & FAULT_FLAG_RETRY_NOWAIT) 485 goto out_up; 486 goto out; 487 } 488 if (unlikely(fault & VM_FAULT_ERROR)) 489 goto out_up; 490 491 /* 492 * Major/minor page fault accounting is only done on the 493 * initial attempt. If we go through a retry, it is extremely 494 * likely that the page will be found in page cache at that point. 495 */ 496 if (flags & FAULT_FLAG_ALLOW_RETRY) { 497 if (fault & VM_FAULT_MAJOR) { 498 tsk->maj_flt++; 499 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 500 regs, address); 501 } else { 502 tsk->min_flt++; 503 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 504 regs, address); 505 } 506 if (fault & VM_FAULT_RETRY) { 507 if (IS_ENABLED(CONFIG_PGSTE) && gmap && 508 (flags & FAULT_FLAG_RETRY_NOWAIT)) { 509 /* FAULT_FLAG_RETRY_NOWAIT has been set, 510 * mmap_lock has not been released */ 511 current->thread.gmap_pfault = 1; 512 fault = VM_FAULT_PFAULT; 513 goto out_up; 514 } 515 flags &= ~FAULT_FLAG_RETRY_NOWAIT; 516 flags |= FAULT_FLAG_TRIED; 517 mmap_read_lock(mm); 518 goto retry; 519 } 520 } 521 if (IS_ENABLED(CONFIG_PGSTE) && gmap) { 522 address = __gmap_link(gmap, current->thread.gmap_addr, 523 address); 524 if (address == -EFAULT) { 525 fault = VM_FAULT_BADMAP; 526 goto out_up; 527 } 528 if (address == -ENOMEM) { 529 fault = VM_FAULT_OOM; 530 goto out_up; 531 } 532 } 533 fault = 0; 534 out_up: 535 mmap_read_unlock(mm); 536 out: 537 return fault; 538 } 539 540 void do_protection_exception(struct pt_regs *regs) 541 { 542 unsigned long trans_exc_code; 543 int access; 544 vm_fault_t fault; 545 546 trans_exc_code = regs->int_parm_long; 547 /* 548 * Protection exceptions are suppressing, decrement psw address. 549 * The exception to this rule are aborted transactions, for these 550 * the PSW already points to the correct location. 551 */ 552 if (!(regs->int_code & 0x200)) 553 regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16); 554 /* 555 * Check for low-address protection. This needs to be treated 556 * as a special case because the translation exception code 557 * field is not guaranteed to contain valid data in this case. 558 */ 559 if (unlikely(!(trans_exc_code & 4))) { 560 do_low_address(regs); 561 return; 562 } 563 if (unlikely(MACHINE_HAS_NX && (trans_exc_code & 0x80))) { 564 regs->int_parm_long = (trans_exc_code & ~PAGE_MASK) | 565 (regs->psw.addr & PAGE_MASK); 566 access = VM_EXEC; 567 fault = VM_FAULT_BADACCESS; 568 } else { 569 access = VM_WRITE; 570 fault = do_exception(regs, access); 571 } 572 if (unlikely(fault)) 573 do_fault_error(regs, access, fault); 574 } 575 NOKPROBE_SYMBOL(do_protection_exception); 576 577 void do_dat_exception(struct pt_regs *regs) 578 { 579 int access; 580 vm_fault_t fault; 581 582 access = VM_ACCESS_FLAGS; 583 fault = do_exception(regs, access); 584 if (unlikely(fault)) 585 do_fault_error(regs, access, fault); 586 } 587 NOKPROBE_SYMBOL(do_dat_exception); 588 589 #ifdef CONFIG_PFAULT 590 /* 591 * 'pfault' pseudo page faults routines. 592 */ 593 static int pfault_disable; 594 595 static int __init nopfault(char *str) 596 { 597 pfault_disable = 1; 598 return 1; 599 } 600 601 __setup("nopfault", nopfault); 602 603 struct pfault_refbk { 604 u16 refdiagc; 605 u16 reffcode; 606 u16 refdwlen; 607 u16 refversn; 608 u64 refgaddr; 609 u64 refselmk; 610 u64 refcmpmk; 611 u64 reserved; 612 } __attribute__ ((packed, aligned(8))); 613 614 static struct pfault_refbk pfault_init_refbk = { 615 .refdiagc = 0x258, 616 .reffcode = 0, 617 .refdwlen = 5, 618 .refversn = 2, 619 .refgaddr = __LC_LPP, 620 .refselmk = 1ULL << 48, 621 .refcmpmk = 1ULL << 48, 622 .reserved = __PF_RES_FIELD 623 }; 624 625 int pfault_init(void) 626 { 627 int rc; 628 629 if (pfault_disable) 630 return -1; 631 diag_stat_inc(DIAG_STAT_X258); 632 asm volatile( 633 " diag %1,%0,0x258\n" 634 "0: j 2f\n" 635 "1: la %0,8\n" 636 "2:\n" 637 EX_TABLE(0b,1b) 638 : "=d" (rc) 639 : "a" (&pfault_init_refbk), "m" (pfault_init_refbk) : "cc"); 640 return rc; 641 } 642 643 static struct pfault_refbk pfault_fini_refbk = { 644 .refdiagc = 0x258, 645 .reffcode = 1, 646 .refdwlen = 5, 647 .refversn = 2, 648 }; 649 650 void pfault_fini(void) 651 { 652 653 if (pfault_disable) 654 return; 655 diag_stat_inc(DIAG_STAT_X258); 656 asm volatile( 657 " diag %0,0,0x258\n" 658 "0: nopr %%r7\n" 659 EX_TABLE(0b,0b) 660 : : "a" (&pfault_fini_refbk), "m" (pfault_fini_refbk) : "cc"); 661 } 662 663 static DEFINE_SPINLOCK(pfault_lock); 664 static LIST_HEAD(pfault_list); 665 666 #define PF_COMPLETE 0x0080 667 668 /* 669 * The mechanism of our pfault code: if Linux is running as guest, runs a user 670 * space process and the user space process accesses a page that the host has 671 * paged out we get a pfault interrupt. 672 * 673 * This allows us, within the guest, to schedule a different process. Without 674 * this mechanism the host would have to suspend the whole virtual cpu until 675 * the page has been paged in. 676 * 677 * So when we get such an interrupt then we set the state of the current task 678 * to uninterruptible and also set the need_resched flag. Both happens within 679 * interrupt context(!). If we later on want to return to user space we 680 * recognize the need_resched flag and then call schedule(). It's not very 681 * obvious how this works... 682 * 683 * Of course we have a lot of additional fun with the completion interrupt (-> 684 * host signals that a page of a process has been paged in and the process can 685 * continue to run). This interrupt can arrive on any cpu and, since we have 686 * virtual cpus, actually appear before the interrupt that signals that a page 687 * is missing. 688 */ 689 static void pfault_interrupt(struct ext_code ext_code, 690 unsigned int param32, unsigned long param64) 691 { 692 struct task_struct *tsk; 693 __u16 subcode; 694 pid_t pid; 695 696 /* 697 * Get the external interruption subcode & pfault initial/completion 698 * signal bit. VM stores this in the 'cpu address' field associated 699 * with the external interrupt. 700 */ 701 subcode = ext_code.subcode; 702 if ((subcode & 0xff00) != __SUBCODE_MASK) 703 return; 704 inc_irq_stat(IRQEXT_PFL); 705 /* Get the token (= pid of the affected task). */ 706 pid = param64 & LPP_PID_MASK; 707 rcu_read_lock(); 708 tsk = find_task_by_pid_ns(pid, &init_pid_ns); 709 if (tsk) 710 get_task_struct(tsk); 711 rcu_read_unlock(); 712 if (!tsk) 713 return; 714 spin_lock(&pfault_lock); 715 if (subcode & PF_COMPLETE) { 716 /* signal bit is set -> a page has been swapped in by VM */ 717 if (tsk->thread.pfault_wait == 1) { 718 /* Initial interrupt was faster than the completion 719 * interrupt. pfault_wait is valid. Set pfault_wait 720 * back to zero and wake up the process. This can 721 * safely be done because the task is still sleeping 722 * and can't produce new pfaults. */ 723 tsk->thread.pfault_wait = 0; 724 list_del(&tsk->thread.list); 725 wake_up_process(tsk); 726 put_task_struct(tsk); 727 } else { 728 /* Completion interrupt was faster than initial 729 * interrupt. Set pfault_wait to -1 so the initial 730 * interrupt doesn't put the task to sleep. 731 * If the task is not running, ignore the completion 732 * interrupt since it must be a leftover of a PFAULT 733 * CANCEL operation which didn't remove all pending 734 * completion interrupts. */ 735 if (tsk->state == TASK_RUNNING) 736 tsk->thread.pfault_wait = -1; 737 } 738 } else { 739 /* signal bit not set -> a real page is missing. */ 740 if (WARN_ON_ONCE(tsk != current)) 741 goto out; 742 if (tsk->thread.pfault_wait == 1) { 743 /* Already on the list with a reference: put to sleep */ 744 goto block; 745 } else if (tsk->thread.pfault_wait == -1) { 746 /* Completion interrupt was faster than the initial 747 * interrupt (pfault_wait == -1). Set pfault_wait 748 * back to zero and exit. */ 749 tsk->thread.pfault_wait = 0; 750 } else { 751 /* Initial interrupt arrived before completion 752 * interrupt. Let the task sleep. 753 * An extra task reference is needed since a different 754 * cpu may set the task state to TASK_RUNNING again 755 * before the scheduler is reached. */ 756 get_task_struct(tsk); 757 tsk->thread.pfault_wait = 1; 758 list_add(&tsk->thread.list, &pfault_list); 759 block: 760 /* Since this must be a userspace fault, there 761 * is no kernel task state to trample. Rely on the 762 * return to userspace schedule() to block. */ 763 __set_current_state(TASK_UNINTERRUPTIBLE); 764 set_tsk_need_resched(tsk); 765 set_preempt_need_resched(); 766 } 767 } 768 out: 769 spin_unlock(&pfault_lock); 770 put_task_struct(tsk); 771 } 772 773 static int pfault_cpu_dead(unsigned int cpu) 774 { 775 struct thread_struct *thread, *next; 776 struct task_struct *tsk; 777 778 spin_lock_irq(&pfault_lock); 779 list_for_each_entry_safe(thread, next, &pfault_list, list) { 780 thread->pfault_wait = 0; 781 list_del(&thread->list); 782 tsk = container_of(thread, struct task_struct, thread); 783 wake_up_process(tsk); 784 put_task_struct(tsk); 785 } 786 spin_unlock_irq(&pfault_lock); 787 return 0; 788 } 789 790 static int __init pfault_irq_init(void) 791 { 792 int rc; 793 794 rc = register_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); 795 if (rc) 796 goto out_extint; 797 rc = pfault_init() == 0 ? 0 : -EOPNOTSUPP; 798 if (rc) 799 goto out_pfault; 800 irq_subclass_register(IRQ_SUBCLASS_SERVICE_SIGNAL); 801 cpuhp_setup_state_nocalls(CPUHP_S390_PFAULT_DEAD, "s390/pfault:dead", 802 NULL, pfault_cpu_dead); 803 return 0; 804 805 out_pfault: 806 unregister_external_irq(EXT_IRQ_CP_SERVICE, pfault_interrupt); 807 out_extint: 808 pfault_disable = 1; 809 return rc; 810 } 811 early_initcall(pfault_irq_init); 812 813 #endif /* CONFIG_PFAULT */ 814 815 #if IS_ENABLED(CONFIG_PGSTE) 816 void do_secure_storage_access(struct pt_regs *regs) 817 { 818 unsigned long addr = regs->int_parm_long & __FAIL_ADDR_MASK; 819 struct vm_area_struct *vma; 820 struct mm_struct *mm; 821 struct page *page; 822 int rc; 823 824 switch (get_fault_type(regs)) { 825 case USER_FAULT: 826 mm = current->mm; 827 mmap_read_lock(mm); 828 vma = find_vma(mm, addr); 829 if (!vma) { 830 mmap_read_unlock(mm); 831 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP); 832 break; 833 } 834 page = follow_page(vma, addr, FOLL_WRITE | FOLL_GET); 835 if (IS_ERR_OR_NULL(page)) { 836 mmap_read_unlock(mm); 837 break; 838 } 839 if (arch_make_page_accessible(page)) 840 send_sig(SIGSEGV, current, 0); 841 put_page(page); 842 mmap_read_unlock(mm); 843 break; 844 case KERNEL_FAULT: 845 page = phys_to_page(addr); 846 if (unlikely(!try_get_page(page))) 847 break; 848 rc = arch_make_page_accessible(page); 849 put_page(page); 850 if (rc) 851 BUG(); 852 break; 853 case VDSO_FAULT: 854 case GMAP_FAULT: 855 default: 856 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP); 857 WARN_ON_ONCE(1); 858 } 859 } 860 NOKPROBE_SYMBOL(do_secure_storage_access); 861 862 void do_non_secure_storage_access(struct pt_regs *regs) 863 { 864 unsigned long gaddr = regs->int_parm_long & __FAIL_ADDR_MASK; 865 struct gmap *gmap = (struct gmap *)S390_lowcore.gmap; 866 867 if (get_fault_type(regs) != GMAP_FAULT) { 868 do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP); 869 WARN_ON_ONCE(1); 870 return; 871 } 872 873 if (gmap_convert_to_secure(gmap, gaddr) == -EINVAL) 874 send_sig(SIGSEGV, current, 0); 875 } 876 NOKPROBE_SYMBOL(do_non_secure_storage_access); 877 878 #else 879 void do_secure_storage_access(struct pt_regs *regs) 880 { 881 default_trap_handler(regs); 882 } 883 884 void do_non_secure_storage_access(struct pt_regs *regs) 885 { 886 default_trap_handler(regs); 887 } 888 #endif 889